Systematic Literature Review on AI and Non-AI guided Image Encryption Techniques

The proposed research is a systematic literature review of image encryption methods, comparing non-AI strategies with those of the emerging AI-driven. Non‑AI techniques include classical cryptosystems adapted to images and a wide range of chaos‑based permutation–diffusion schemes that exploit sensitivity to initial conditions and ergodicity for secure pixel scrambling and intensity modification. Such techniques usually provide security properties that are well understood and computationally complex that are relatively low but are not as adaptable to a variety of image types and attack models that change over time. The review then looks at AI-directed methods which involve applying computational intelligence and deep learning to different phases of the encryption pipeline, including generation of keys, optimization of parameters, and end-to-end learned ciphers. Adaptive key spaces and search capabilities are offered by neural networks, genetic algorithms, fuzzy logic, and strong statistical security measures and application-specific performance are demonstrated by CNN and GAN-based schemes in particular fields of medical image protection. In both types, reported metrics (i.e. information entropy, NPCR, UACI, pixel correlation, and computational cost) are synthesized in the review, revealing trade-offs between security and efficiency and complexity of implementation. The analysis finds open issues with non-AI methods in regard to key management and resistance to advanced cryptanalysis, and with AI methods in regard to explainability, formal security certificates, data dependence, and resistance to adaptive attacks.